ÆRA LLM: Artificial Intelligence for Enterprises

Many companies face a paradox, t hey want generative AI but don’t intend to share their data—like driving with your foot on the accelerator and the brake at the same time. They seek cost predictability, yet end up with variable charges that turn every billing cycle into financial roulette.
To meet these needs, we built ÆRA: an open-weights Large Language Model, deployable on-premises and free from token-based variable costs, purpose-built for enterprises with a focus on factuality and answer reliability.

The ÆRA approach: factual and transparent

With 4 billion parameters, ÆRA doesn’t compete on size but on effectiveness: it’s engineered to be fast and accurate. Its open-weights architecture means transparency: the model can be inspected, modified, and trained on your proprietary data until it becomes an expert in your domain.
ÆRA’s core strength lies in its architecture oriented toward factuality:

• Rigorous context adherence: the model answers only when it has sufficient information in the context or via called tools, avoiding the “hallucinations” typical of generalist models.
• Native function calling: it dynamically queries external knowledge bases and internal document repositories, fully bypassing pre-training knowledge when necessary.
• Source traceability: every answer can be traced back to its information source, ensuring verifiability and compliance.
• Structured outputs: it excels at following and validating schemas (e.g., JSON Schema or Pydantic models), producing responses ready for integration into pipelines and APIs with no post-processing.
  

The model is optimized for Italian and English and It will be available on Hugging Face in the next few days. It’s the first in a series on our roadmap.

We’re not starting from scratch.
In recent years, we’ve designed and put into production AI models and platforms for enterprises—from fine-tuning on proprietary data to orchestration and security in regulated environments.
ÆRA is the synthesis of this experience: a local (on-prem) and transparent (open-weights) model, designed for business use.

I know what you're thinking: do we really need another LLM? I asked myself the same question. Not necessarily. What companies need is someone with deep and cross-functional expertise in developing projects that generate real impact.
On the other hand, if it's true that 95% of AI projects don't have a positive impact on the bottom line (The GenAI Divide: State of AI in Business 2025 – MIT), it's probably because it's easy to convince yourself you "know what to do" with AI, but without rigor you're flying blind.
Developing your own series of models means getting your hands dirty and understanding the engine under the hood: it's not always necessary, but when it is, it makes all the difference. It's the distance between assembling IKEA furniture and designing custom furnishings.

That’s the AND EMILI way: prioritize facts over words.

 
Where ÆRA makes the difference

• Management of specialized knowledge bases: when you need to operate in highly vertical, domain-specific knowledge areas.
  
  Imagine assessing the creditworthiness of Italian SMEs. A generalist model knows financial statements and indicators (ratios); it doesn’t know your predictive signals, the sector correlations discovered in the field, the weak signals that precede defaults in your portfolio. ÆRA, by contrast, can be specialized on your data and processes until it becomes the best expert on your market.
• Data sovereignty and compliance: your data remains yours. No sharing with third parties; full support for GDPR compliance, any sector-specific regulations (ECB/EBA), and DORA requirements on digital resilience. Complete traceability of operations, segregation of sensitive data—everything needed to pass the toughest audits.
• Predictable TCO: for high-volume scenarios, ÆRA eliminates the token-based variable cost component.
  
  In addition, ÆRA can be further refined through fine-tuning, allowing you to optimize performance for specific use cases and achieve higher accuracy in the desired application domain.

Integration with the AND EMILI Cognitive Platform
ÆRA integrates natively with the AND EMILI Cognitive Platform, enabling the orchestration of specialized virtual agents that:

• For Marketing: converse in your tone of voice, powered by proprietary content, delivering dashboards of real insights (topics, sentiment, trends).
• For IT: integrate with your existing infrastructure (CRM, automations, digital channels) while keeping full control over technological evolution.

Try ÆRA